Methods of three dimensional reconstruction for personalised prosthetics and comparison using preserved anatomical landmarks

Abstract

Personalised prosthetics requires anatomical models whose dimensions and local shape remain consistent with the observed patient anatomy. Medical images and imported surface models differ in their coordinate systems, units and completeness, so a geometric discrepancy cannot be interpreted before registration and input verification. This study compares geometric, statistical and learned approaches and describes an implemented workflow for comparing a reference surface with a patient surface. The workflow preserves the physical dimensions of the patient model, estimates a similarity transformation from landmarks on the preserved region, computes distances to the observed surface and checks whether the available information supports the selection of prosthetic candidates. A controlled computational experiment uses a synthetic geometric phantom with a known missing component. Four registration variants are compared under three levels of landmark noise, with thirty repetitions at each level. At a landmark noise standard deviation of half a millimetre, the preserved-landmark method gives a mean target registration error of approximately forty-five hundredths of a millimetre, compared with approximately one and eighty-four hundredths of a millimetre for bounding-box scaling followed by rigid registration. The relative reduction is approximately seventy-five per cent in this specific experiment. The result illustrates how estimating scale from preserved anatomy can reduce the bias caused by a missing component. It does not establish clinical superiority or superiority over trained statistical or neural shape models. An extended objective with segmentation, local deformation and a defect mask is formulated as a research direction and is distinguished from the evaluated implementation. The system produces geometric measurements and candidate recommendations for specialist review; it does not automatically diagnose an injury or prescribe a prosthesis.